Although the desktop is a big market and will continue to be one for some time, the lack of growth means a likely shift from research to development (to make things cheaper) as the competition reduces the gross margins, but new research doesn't pay back (it's good enough). The platform will probably die a slow death as new research money (and they people that chase that money) move on to the shiny new stuff and as vendor consolidate to reduce overhead and competition, the number of product offerings will shrink.
If you want to compare this with the mainframe, there are only really 2 types of mainframes today. IBM (zOS) and clusters. Although people often talk about servers and super-computers as mainframes, they really aren't the same thing. Although once a highly vibrant multi-vendor industry, modern mainframes have specialized to the point where they are really just encrypted transactional database processing machines provided mostly by one company (IBM 90% market share). It may be a multi-billion dollar industry, but it's mostly vertically integrated and single sourced. I think if the PC business morphed back into this model because of lack of growth, I doubt anyone would recognize this as the desktop that they know and love anymore...
I think a better analogy for the desktop is the wristwatch business. Although it might appear that there are many watch manufactures, most vendor just buy a movement from a handful of vendors. This is pretty much how the desktop is now (small number of ODMs, selling standard stuff to companies that repackage).
For a while, the end-user price of a watch seemed to be going down and there are many available options, but more recently the price of a decent watch is instead it is going up. Why? Because instead of being a growth industry, less people are buying watches (because everyone has a time-telling cell-phone) and the low-cost players chased out all the bulk of the industry and the remaining players are selling jewelry that happens to also be a watch. The movements in these low-end products haven't really changed in years. Of course there are always a few niche players that are more vertically integrated or buy high-end movements direct from other vertically integrated vendors and craft them into custom products, but you pay through the nose to buy them because there is no down-market to create a volume business as the velocity of innovation slows.
In the watch biz, high-end movements today don't get picked up by the low-end folks after a few years, the low-end movements are completely optimized differently for price and manufacturablity, the old high-end stuff just gets end-of-life/discontinued because the parts were never produced in enough volumes for a low-end volume business and the high-end folks don't want to canabalize their own business by dumping their old high-end stuff in the discount bin.
I think the PC desktop business has already bifurcated into this today. If you want a generic vanilla low-end platform where you can install linux, it will probably exist for a long time. If you want something high-end, you will likely pay the "niche-tax" for it. Who knows, it might track the watch business as desktop usage declines, but unless a low-end "jewelry-like" segment emerges, the low-end may just evaporate because of low-demand. I think many folks feel the tablet will emerge as the "jewelry" segment so in this case the low-end of the desktop business might be doomed. In which case, it might still be a multi-million dollar business, but each PC will be at least $2,000 (ironically, the same as the low-side of the high-end retail watch business) instead of under $500 (the average retail price of a "jewelry" watch before it hits the discount bin).