In general you are right, though in this case he missed a pure 8 ply combination. I believe it was a Horizon problem. He didn't scan the whole 8 ply depth and missed this variation.
It is to me not a pure endgame mistake which you typically recognise in an voluntary piece exchange into a lost endgame.
An 8 bit program at no more then 4Mhz typically considers 7 up to 8 ply @ tournament play (3 minutes a move) The progam uses the shannon A Strategy, meaning he will calculate the whole tree.
This takes time as the possibilities are on average x40 for every ply added.

This is also a reason why I have such a bad score against the Turbostar (5 to 10%) . I believe the Turbostar 432 KSO has a A & B-strategy program and is a real frustration for me. It scans the tree for a few ply (3 to 4 A Strategy) and then makes a few deep dive considerations on the most promising variations like humans do (B-Stategy) in an attempt to eliminate the calculations of stupid/non-promising move considerations.
Of course a sacrifice combination can be quickly overlooked (like a queen sacrifice to mate) as initially catalogued unpromising , therefore the whole variation may be quickly dropped.
Either choice was a programmer choice back in the 70'ies , 80'ies and early 90'ies.

https://chessprogramming.wikispaces.com/Type+A+Strategy
https://chessprogramming.wikispaces.com/Type+B+Strategy