From a pure protocol point of view the SPF record does have one major advantage over TXT and that is in the use of wildcard records.
In short a wildcard on a TXT record for SPF is going to have impact on every other scheme that overloads TXT, of which there are many. SPF does have a mechanism to resolve the ambiguity but that does not stop the record sets from getting to be larger than will fit in DNS UDP responses.
This has not been much of a problem because most of the other TXT overloads don't have a use for a wildcard. There is not much point in a wildcard DANE record. But it could be an issue.
So keeping SPF records does actually have a utility since it allows wildcarded records to be isolated from other protocols and avoiding the wildcard record set bloat issue.
The criteria to use to decide is not the proportion of SPF records published as TXT vs SPF but what the validators look for. If at this point there is little to no takeup by the validators for the SPF RR then it is time to call time on a failed experiment. If there was a significant support base for SPF RR validation then it is a harder call.