They're not exploring a research space, they're exploring a space where the options are incredibly well known, and they're jumping from solution to solution, even after they appear to find a success.
If they have enough resources, they can afford to try all options in parallel. Wasteful, but although it may seem strange, it isn't like programming language implementation engineering is expensive compared to data centers.
yeah, god forbid they feel that their initial success could be improved upon; They should put down their tools as soon as they get their first version that works