Long thread that I didn't read entirely, but here's my experience:
PIC has always had the best hardware features for the cost, but an awful CPU architecture as far as a C compiler is concerned. C was not the original design goal, and it seems like they're stuck with what they already had and won't change. So to use it well requires a proprietary compiler that is quite expensive for the full version. There is a free one, but it produces assembly so bad that it's been accused of purposely sandbagging to make the paid versions look even better. (actually, if you really know how a compiler works, you'll understand that the free version really is just the translation stage as-is with nothing after it, and that the optimizer really is that good) There are also community-supported compilers, but they don't get a lot of attention, and so they're still not as good as the proprietary one. It's too far different from a "standard" architecture for GCC to be modified to work.
So PIC is excellent in value for hardware, but bad for development. Not really a problem if you develop once and then dump the same binary into millions of chips.
AVR (both before and after MCP bought it) is exactly what GCC expects, so there's no reason to use MCP's version of it. Just get the free AVR-GCC and move on. It also runs 1 instruction per clock instead of PIC's 1 instruction per 4 clocks, but it requires more instructions to do some common things, so the benefit isn't quite as much as you might think. Most of the practical benefit is a boatload of RAM, and a NOP is 1/4 the time on AVR as it is on a PIC, so you can busy-wait more precisely. But it doesn't have as much on-chip hardware as a similarly-priced PIC. I see no reason why they can't put a PIC's peripherals around an AVR core for the best of both worlds...except that they wouldn't be able to sell their expensive compiler anymore.
So AVR has a better CPU, which makes it easier to use as your first microcontroller, but so far isn't paired with as much hardware for the price.
---
Ultimately, Microchip is out to make a profit. They can't operate as a charity, or they'd go bankrupt and cease to exist. Can they make all of their profit on hardware and let hobbyists use their full-version software for free? Perhaps, but I don't think they're focused so much on hobbyists. Hobbyists are too low-volume to be profitable to a chip manufacturer, and if you offer something to them for free, then the large corporate customers can get it for free too. (a "non-commercial" clause in the license isn't as effective as you might think)
It'll be interesting to see what the next generation does. The idea that those who learn on a given system will later decide to use it in production is very real, and those who are just starting to learn will choose the one that's most accessible to zero knowledge and almost zero budget in single quantities. So a manufacturer can't just ignore those people. Sure they're not profitable now, but they certainly will be if they're not driven away!
---
One more note:
I see people complaining about the PicKit. I have one, and haven't had any issues so far, but I also found a free programmer that uses the GPIO pins of a Raspberry Pi:
https://wiki.kewl.org/dokuwiki/projects:pickleI've only used that for one project so far, which is a permanent connection between a Pi and a fairly new PIC, but it works perfectly! Together with SDCC in Code::Blocks, I have a completely free toolchain that runs entirely on the Pi. Nothing at all from MCP except the chip itself. (SDCC still has some significant shortcomings, but it's a lot better than MCP's free thing!)