Increasing the input size does seem to have an effect on the relative performance of the different machines. This can be observed for most of the applications. Only VL2MV which is computation intensive and doesn't use much memory has uniform relative performance.
A lot of exceptions to the ``rule'' (or rather misconception) that faster is better were found. For the application SIS, it was found that the MIPS R4400 machine (60MHz) was performing better than the Sparc20 (100MHz). For VIS with non-local memory accesses, Sparc20 performed very well because of its very fast 256K second level cache compared to the Alpha 21064 machine with higher clock speeds. But with local memory accesses, the advantage of fast second level cache decreased and the greater clock speed of the Alpha 21064 (182MHz) made it faster. In general, we found that for CAD applications with inputs that fit in the main memory, performance is a complex function of clock speed and cache organization.