They're very special cases, due to an extremely unusual circumstance called Moore's Law.
Maybe I give you more credit than you deserve - but naming a phenomena. like "Moore's law" does not explain anything. The question is why some prices decrease while others are more or less static in "real terms".
I can't think of anything that's CNC'd on a production car? A couple small parts in the engine? Everything is pretty much cast and stamped.
What are you thinking about ? Every moving part in your engine is machined - the pistons, piston pins, cylinders or inserts , sleeves, cams, cranks, mating surfaces, valves, the bearings even the bolts. Typically the cam and crank bearings are inserts - machined separately and inserted. Sheesh - buy a clue or read a book ! Precision casting isn't close to the tolerances needed. Stamping (which is neither machining nor casting) is used for oil pans and cover plates but not part of the drive system. You certainly have well above 1 sq.meter of machined surfaces in the drive train.
Actually, from about 1980 through around 2000, the cost of a high-end desktop PC stayed pretty much constant -- you just kept getting more and more for that price. It's only in the last few years that the actual price has dropped significantly.
Funny - a friend was showing off a 1980s PC catalogue and the prices haven't changed much - tho the capacities have increased radically.
===
It's certainly true that high density ICs (CPUs and memory for example) have decreased in price per transistor or price per performance or capacity, but why ?
There is a vast capital cost in setting up an IC "fab" manufacturing line (several billion today), it takes several years, the engineering costs to design a modern CPU is huge as well, but once the line is in production the recurring costs (labor, chemicals, high purity silicon wafers) are tiny compared to the value of the product. So *IF* you can make good use of that fab you can make decent money. If you can't - if the competition tops your products - then you lose on that huge infrastructure investment.
So the interesting thing is that decreasing line size (the 65nm, 45nm or 32nm feature size of an IC surface structure) both increases the number of chips per recurring cost (chips/wafer, chips/unit laborand) also increases the potential chip speed and decreases the chip poewr requirement - but also at least initially increases the part failure rate. So there is an interesting balancing act; smaller is better except you must account for more failures.
Of course bonding out ICs and making/assembling circuit boards is currently labor intensive and is farmed out to 3rd world countries.
Back to Moore's law - every time you decrease the linewidth by ~30% you can double the number of parts and improve their characteristics. It's a very rare thing that these "incremental" improvements can cause such an improvement in production and quality.
Disk storage has also increased radically in capacity/$ since it has a similar economics. If you can polarize smaller magnetic domains on the disk and space tracks closer - then you get higher capacity AND faster prefromance as a adirect result of this one-dimensional improvement.
Back in the 1980s there was popular chatter that an ideal workstation (new concept) would have the "3Ms" 1 MIPS, 1 Megabyte and 1 Megapixel (sometimes 10MB of disk was added to the list). I just upgraded a PC (~$400) to ~7GIPs, 4096Megabytes, ~1.5Megapixel, and 1000000MB of disk. So it's notable that this common system beats the 3M by factors of 7000,4000,1.5 (and disk by 100,000). Clearly display technology has languished while IC and disk have improved tremendously.
More typically in production (cars for example and perhaps computer displays) there are many disparate parts and many separate steps and each one has separate price/production issues. Someone mentioned a car engine is ~$3k - so even driving the engine price to zero has only a modest impact on price. Improving one of these has only a small impact. The fact that a semiconductor wafer is a homogeneous surface and tho' the processing is detailed and occurs in several common steps. It's an ideal situation to look for incremental optimizations. Similarly for disk surfaces and heads.
So one way that we get improvement in industrial production of these complex parts is to simplify the design ... use standard modules, stamped or cast pasts wherever possible, remove assembly steps, reduce the number of parts. Reduce the labor costs. We see this trend all the time. If you look at a electric toothbrush, vacuum cleaner a home furnace fan motor or a hose spigot you'll see the design consists of an minimal number of parts. If you examine the manufacture of a KB or mouse from the 1990s vs today the simplification is clearcut. Modern cheap coffee makers have far fewer simpler parts than the pre- Mr.Coffee units. The simple modular design means the potential for repair is greatly reduced. Repair consists of replacing a whole module or the whole unit. This is one of the trade-offs that makes it somewhat undesirable to do larger scale modularize of high expense items like an auto. Current auto design has too many parts - too complex to become cheap.
Electric vehicles drive train and braking have considerably better potential for modularization than IC engines, transmissions with 2/4 WD and independent braking. Even moreso for fuel cell technology.
So ultimately, IMO, the costs of production are driven by complexity and non-homogeneity (of a sort) which results in high production and assembly costs. I'm sure there is a way to caprure the issue in information theoretic language, but it escapes me for the moment.
Interesting.