Well, they are 16-bits devices with more that 64kb in combined flash+memory, so that always requires some special handling.
That is not what I actually meant, I only recall now there was some weird addressing like 3 bytes on every even address in FLASH or so, making a custom bootloader application really a tough job. That addressing was due to some historic reasons and very badly designed.
Ah that. Yes there is something like that, though I can't remember if it is notational (3 bytes in hex to 2 bytes in flash) or not.
What I really like is the high pincount, and the availability of the whole range of peripherals on one part.
That does almost every 32-bit Cortex now, with linear memory addressing and all the powers of the ARM core,
(the dspic if you enable the upper memory area to be ram not flash is linear too, there are compiler options for that. Specially if you have little static initialize data this is a good solution).
CMSIS standard for easy porting between vendors etc.
I don't really care about porting, if you years later have to port because the part is unavailable, reopening development is a problem in all cases. It is not really a decision influencing factor to me, because so easily remedied. If a part is EOL (and that is *RARE* with microchip) If push comes to I drop a couple of hundreds bucks worth of on chips on store, and am good for another 5-8 years.
But most important are the peripherals, I looked not one year ago, and I didn't even come close, specially motor control. How many Cortex parts e.g. have quadrature encoder ? The LPC range looked better than it did 5 years ago though.
For my part, never needed more than 100 pins (which e.g. TI Stellaris or NXP LPC parts offer for comparable prices).
I mostly use the 144pin part. The generic board is used the most, because we don't turn over that many, (tens an year) it was cheaper to design in the biggest part (well second biggest, but it only differs in flash amounts) than have multiple boards to save costs. The P&P for extra prints was already more expensive.
Generic boards however also in general like remappable I/O (which was also was a bit scarce in LPC iirc)
The CPU still isn't the most expensive part on the boards, high speed optocouplers (for quadrature) usually are.
Admitted we also use the 144pin part because it also has the highest number of remappable I/O.
That said there are also points with concern. CPU speed is not really an issue, but recently we added a IP stack IC, and SPI speeds could be higher. (something like 15mbit for the one hardwired to pins, 10-8 for the remappable ones).
Luckily it is all under DMA (even command and control of the IP stack IC, one of the reasons to select it, e.g. the 32-bit PIC do much better there, but are motorcontrol is not available, and we don't have a major problem to begin with)
Maybe a major redesign will go a different direction, but this design has quite some mileage yet, and not really problems (incremental revision from F to E about two years ago)
It's never good to view a cpu family through eyes shaped only by other vendors. It makes it easy to look over the actual strengths.
Microchip had free tools as the first vendor, that was their only strength. But some other ppl have different opinion as you, as can be seen e.g. here: http://www.microchip.com/forums/m601857-p2.aspx
I know that is the popular opinion, but IMHO it is wrong or at the very least oversimplified. It all depends on what you want. Many of the popular opinion consider raw ALU speed to be the most important. Some from the other camp bitbanging speed (dspic can toggle a pin at half the CPU freq).
Mine is probably colored by motor control (and then specially quadrature) needs. Some people prefer one part (we mostly, except for the (TCP/)IP engine) , some are more amenable to add additional parts.
Davy Jones has the right of it:
https://www.youtube.com/watch?v=DBftApUQ8QI