That's not accurate. Even today many people run nodes supporting different features. Even before the softforks, nodes upgraded to support the features if the softfork might be activated. The problem right now is that Core reviewers are the gatekeepers for consensus implementation. For a somewhat related example, take Bitcoin Knots. It is essentially a filter that still respects the most proof-of-work chain whatever that might be. The point of it is that if a majority of people adopted it, miners would be forced to update their consensus rules for fear of not having their blocks propagated. Let's say a worthy alternative implementation gained majority popularity. Would consensus proposal review and reference implementation remain on Core?