There is no length/1 optimization. Capture length/1 instead of write code every time.

bench.erl

-module(bench).
-export([run/0]).
-define(SEED, {1393,433335,396007}).
-define(LENGTH, trunc(math:pow(2, 14))).
-define(TIMES, 3).
-include_lib("eunit/include/eunit.hrl").

gen() ->
    random:seed(?SEED),
    lists:map(fun(_) -> random:uniform() end, lists:duplicate(?LENGTH, 0)).

length_every_time(Nums) ->
    [N / length(Nums) || N <- Nums].

length_one_time(Nums) ->
    Length = length(Nums),
    [N / Length || N <- Nums].

%% 7> bench:run().
%% src/bench.erl:22:<0.38.0>: EveryTime = [602406,692627,1232612]
%% src/bench.erl:24:<0.38.0>: OneTime = [1823,947,2815]
%% ok
run() ->
    Nums = gen(),
    EveryTime = lists:map(fun({T, _}) -> T end, [timer:tc(fun length_every_time/1, [Nums]) || _ <- lists:seq(1, ?TIMES)]),
    ?debugVal(EveryTime),
    OneTime   = lists:map(fun({T, _}) -> T end, [timer:tc(fun length_one_time/1, [Nums])   || _ <- lists:seq(1, ?TIMES)]),
    ?debugVal(OneTime),
    ok.

My environment:

--- erlang
Erlang R16B02 (erts-5.10.3) [source] [64-bit] [smp:4:4] [async-threads:10] [hipe] [kernel-poll:false]

Eshell V5.10.3 (abort with ^G)

--- PC
MacBook Air
Processor: 1.7 GHz Intel Core i7
Memory: 8 GB 1600 MHz DDR3