This is not a direct answer to your question, but it will be helpful. I happen to miss too the playing style of the early chess computers. Those machines ran amazing algorithms on incredibly poor hardware resources (usually 128 or 256 bytes RAM and a 2 to 16 Mhz, 8-bit processors with limited instruction sets, which would make anyone laugh when compared with the cheapest smartphone of today). The resulting game style at a few seconds per move was aesthetically decent chess (rooks placed in open lines, knights towards the center and so on) that tried to use the machine resources as good as possible (instead of being purposely dumbed down like modern engines) but with the so-called "horizon effect". This allowed the attentive human player to beat the computer by means of combinations whose final result was above the horizon of the few plyes the computer was able to examine, e.g. you could trick the machine with a sacrifice for instance, or you could detect a weakness in an otherwise apparently sound position.
Yes, as a result, those chess computers played what many people like me find more human than all the PC chess software that followed. With "modern" software (meaning PCs since ~1990) it is very noticeable when randomly stupid moves are inserted in an otherwise strong play, or when the program plays something ugly with not much sense positionally that does not win you immediately but also does not allow to be beaten with tactics. There is no way you can beat modern software by being tactically strong, that has never again happened since the 90s. It is incredible that almost no one has paid any attention to this, but the solution was very simple: limiting the calculation speed of engines until matching those of the early devices. Without altering anything else: no choosing of suboptimal moves, no random noise added to the evaluation function or anything. Just let the algorithm get the most out of a position but set a very low limit to the number of positions evaluated per second and you have it.
It has been done at least once: try the open source UCI engine Rodent II by Pawel Koziol. Set the parameter NpsLimit at something between 40 and 100 and allow it between 5 and 20 seconds per move. It really plays like those old chess computers.
I am unaware of other engines that implement that possibility. Some engines limit the search depth (Rybka) but the response is nearly instantaneous. Others emulate a slow response but it is not clear what they are doing (looking at the playing style I suspect Ufim might be following this approach but it is not explicit). Others choose suboptimal moves as said (Stockfish), but nearly any author has thought of implementing the full chess knowledge and positional evaluation at arbitrarily low speed, and I hope the idea is copied by many developers. Playing against such algorithms with horizon effect is fun, and it really gives you the impression of a fight against an opponent that is trying to do his/her best, instead of a mighty opponent that is playing sub-optimally on purpose. I think that is what your question is about.
