The DS89C4x0 types are probably plenty fast enough for your application, since they generate one instruction cycle for each clock cycle, unlike the "old" 805x types that divide the clock by 12, or the Atmel parts that divide the clock by 4, or even the other Maxim/Dallas parts of which some divide the clock by 3. These guys have 1Kb of on-board SRAM, in addition to the standard 128 bytes, and they can map it such that it resides in either program or data memory. Further, they contain 16, 32, or 64KB of FLASH, which (except on the 89C420) can be programmed "on-the-fly." They can, under software control, change the rate at which the core divides down the clock, thereby reducing speed/power product, and they can do lots of other things not common to the 80C3xx, 80C4xx, or other parts of Dallas origin. All these guys have two data pointers, so moving things around in memory is pretty easy, AND the memory map can be quite flexibly fiddled with under software control.

They also have some external memory addressing capabilities, called paged addressing, which saves a clock tick in accessing external memory. All this is not particularly interesting if you don't have performance-critical tasks to perform, but if you do, you can see whether they can help you.

I've not fiddled with other than the 89C420 from this series, but, the sheets seem to suggest that the 89C450, with 64KB of internal FLASH, can access that 64KB as well as an external 64KB, AND an external 64KB of data memory/peripheral space as well, and that's all on a standard 8052 DIP-40 or PLCC-44 footprint. With the aid of a small CPLD to latch/buffer the addresses and data, all the page modes are available. Those can also be used with a bus switch or the like, but a fast, small, cheap CPLD is nearly fast enough to allow you to use those modes without having to program (under software control) additional external bus-cycle time, so you can run at maximum warp.

RE