Trust in the abstract isn't inherently transitive, agreed. I'd argue that employing PGP as though it were is misusing the tool (that might well be easier to do than it ought to be, but that's a different conversation).
WoT as realized by PGP seems to me to be a very good tool for manually assessing whether to trust a previously unknown key for someone that a third (untrusted) party sends you.
I remain highly optimistic that some future take on the WoT concept will be advanced enough to tackle trust in general in a distributed manner.