I have two designs in production with the standalone 2515 right now. It's not for me. It's fine for talking and receiving bus network traffic at moderate speeds. But when you get to 750-1000kpbs and need to dynamically filter, add, modify data with math functions on two or more busses it's too damn slow ESPECIALLY over SCI. I suppose I could consider looking at it a parallel mode chip again, but the SJA series only has 1 receive buffer...
The Atmel has no chips with dual can, so it doesn't matter if their 8bit handles XXXX or YYYY better, they are disqualified.
I think ARM's features would be wasted in most of my products.
As far as continuing with PIC goes... The 24F series is nice, but the only 2 can in that series is a 100pin unit, which would be a waste of pins and real estate on my pcbs. Likewise, the DSPic33 has 2 cans on a 64pin unit, but with all the disadvantages of DSP (low current sink/source being the biggie). So if I go with the DSP I'll probably need drivers/fets/buffers to do simple work that I may have been able to use a non-DSP earlier. At that point the 100 pin seems better.
Eh!!!!