Slightly better benchmark access

README.md

@@ -17,7 +17,7 @@ the existing Haskell ecosystem.
 Main features:
 
 1. Very fast
-   ([benchmarks](https://github.com/haskell-effectful/effectful/tree/master/benchmarks)).
+   ([benchmarks](https://github.com/haskell-effectful/effectful/tree/master/benchmarks/README.md)).
 
 2. Easy to use API (if you know how to use the
    [MonadUnliftIO](https://hackage.haskell.org/package/unliftio-core/docs/Control-Monad-IO-Unlift.html#t:MonadUnliftIO)
@@ -74,7 +74,7 @@ It's true that its "effects as classes" approach is widely known and used often.
 However:
 
 - `mtl` style effects are
-  [slow](https://github.com/haskell-effectful/effectful/tree/master/benchmarks).
+  [slow](https://github.com/haskell-effectful/effectful/tree/master/benchmarks/README.md).
 
 - All of most often used monad transformers (except `ReaderT`) used for effect
   implementations are rife with [subtle

benchmarks/README.md

@@ -31,6 +31,11 @@ as that would essentially mean performing whole program specialization.
 
 The code was compiled with GHC 9.2.3 and run on a Ryzen 9 5950x.
 
+*Note:* below results are from a 1000 iteration run. Runs with more iterations
+are not included in the analysis since they are proportionally the same, but can
+be found
+[here](https://github.com/haskell-effectful/effectful/tree/master/benchmarks).
+
 ### Countdown
 
 <img src="https://raw.githubusercontent.com/haskell-effectful/effectful/master/benchmarks/bench_countdown_1000.png">
@@ -48,7 +53,7 @@ Analysis:
    - Its `State` implemented via `IORef` is the fastest of the dynamically
    dispatched effects, but it's worth noting that it's neither properly
    thread-local nor shared as the underlying `IORef` is shared, but can't be
-   safely updated with `get` and `put` from multiple threads.
+   safely accessed with `get` and `put` from multiple threads.
 
 3. `freer-simple` does surprisingly well for a solution that's based on free
    monads.