I had similar concerns when I made the jump from a custom data class (with methods like `set`, `get`, `increment`, `push`, etc) to a transparent `Proxy` around native JavaScript objects. I did some quick benchmarks back then, and concluded that it was actually not meaningfully slower. Modern JavaScript engines are awesome! :-)
Aberdeen is not in js-framework-benchmark yet, but I've done a pull request https://github.com/krausest/js-framework-benchmark/pull/1877 -- By my own testing, performance is similar to something like react (while of course destroying React on time-to-first-paint, bytes transferred, etc). However, this test is not a particular good fit for Aberdeen, as lists in Aberdeen are always sorted by a given key (it maintains an ordered skiplist for the data). The test only requires the easy case: sort by creation time. So Aberdeen is doing some extra work here.
With regard to reactive data transforms: Aberdeen provides some (reactively 'streaming') helper functions for that (`map`, `multiMap`, `partition`, `count`). But these are mostly based on the `onEach` primitive, for reactive iteration. Take a look at the `map` implementation, for instance: https://github.com/vanviegen/aberdeen/blob/a390ce952686da875...
Is past projects, we've been using things like this a lot. Pretty easy and fast!