But why would people choose lodash-fp over lodash (lodash-classic? :-> ) if, as you say, it reduces performance? Why would they choose lodash-fp over Ramda, which values that style API, whereas you've been arguing against it?
lodash-fp is for devs that dig a more functional style.
See the v3 release notes for more info https://github.com/lodash/lodash/releases/tag/3.0.0
lodash optimizes composition by storing metadata for wrapped functions. So while a function may have _.ary, _.curry, and _.rearg applied the original function is only wrapped once.
With lodash-fp devs get the lodash they know, complete with lazy evaluation, in a more functional friendly package.