You're right, in that if one needs SPI, it's probably best to use a variant that provides that. However ... if you really need the features that one MCU provides, like hardware SPI or IIC, or whatever, then there's no point in sticking with another, is there? Maybe, if you use 1000 a week of one, then it's easier to use that one than to try to get purchasing to get you a different one, and run it past FMEA, and component-engineering, etc.

If you need two data pointers, with auto-increment, and you actually do need to use a 1x multiple of the crystal input in Mode 0, (not very common, but it can happen, since you can do some things with Mode 0 that are pretty useful, particularly in synchronous, with external processes, operation) then the Maxim/Dallas part could, in fact, outrun the 100 MIPS MCU that doesn't have those features, because of the numerous things it doesn't have to do. Moreover, since precise synchronization is of interest, Mode 0 is not the only case where this can occur. You could burn more time watching a status bit in unsynchronized operation rather than simply reading or writing a value when you "know" the time is right. Closed-loop operation is nice, but, when time is of the essence, it isn't always necessary, or even possible.

My point remains that it's important to select the MCU with the "right" features for the application. If I needed a MAC, I'd probably want to use the SiLabs part, since I know of no other 805x-family member that has one and can run fast enough for some jobs, but if I want a data pointer that automatically increments, then automatically switches to the "other" data pointer without any additional instructions, well ...

RE