The big downside of Full-RBF is that it effectively made the risk assessment of received transactions harder, something that a service could perform before in order to give each transaction they received a "point" and if it were below a certain threshold they could accept the smaller risk and accept the tx without confirmation.
An example of such system to see was the one offered by the blockcypher explorer called Confidence Factor:
https://www.blockcypher.com/dev/bitcoin/#confidence-factormore users used to trust zero confirms back then, because it was not straight forward in just the node itself to cheat. its required pushtx knowledge and timing and and also the IP addresses of particular nodes to know where to relay and not.
Trusting unconfirmed transactions have always been discouraged in the Bitcoin community, but if we are talking about the early days like 2009 as in your previous comment, then it was a lot easier to perform double spend attacks for two reasons.
First was the smaller number of nodes in total that one could connect to and send a conflicting transaction to each and wait to get one confirmed.
And second was the fact that one could send out a transaction (the actual payment) to the network but mine the next block containing the double spend themselves since the difficulty was very low.