I suspect it was also rational, given the performance trade-offs at the time.
As transistors have become smaller, the caching hierarchy has become more important. A modern register set is a level-0 cache -- i.e. a useful pool of memory that is really close to the the computation units.
In the old days, the communication between components was less of a bottleneck than the speed of the components themselves. So there was less advantage to this. On the other hand, making a CPU simpler by having just one or two registers was more of an advantage than it is now.
As transistors have become smaller, the caching hierarchy has become more important. A modern register set is a level-0 cache -- i.e. a useful pool of memory that is really close to the the computation units.
In the old days, the communication between components was less of a bottleneck than the speed of the components themselves. So there was less advantage to this. On the other hand, making a CPU simpler by having just one or two registers was more of an advantage than it is now.