Is loading bytecode in Luajit script slower than loading strings?

The problem:

I have been testing two types of large files (roughly 20MB).

The first is a string file with around 4000 functions in it like:

function dostuff1()
    <simple lua script>
end

function dostuff2()
    <simple lua script>
end 

And so on. The functions themselves are not overly important, but each one has very specific important simple operations in them.

The second file is simply a bytecode built version of the text file using:

luajit -b testfile1.lua testfile1.out

Then I test each file (load times) for each of them into their own env using:

local env = {}
local chunk = loadfile( filename, 'bt', env )
assert(pcall(chunk()))

The string files are significantly faster to load and process than the bytecode files. Which seems counter to what I would expect. Heres a small sample of comparisions. Note: some files have more complex functions in them than others (hence the time differences).

String files:

'#  File Load+Run: '    'test.tbl1'     '  Time: '      0.72784 '   Size:'      20877534
'#  File Load+Run: '    'test.tbl2'     '  Time: '      0.686193        '   Size:'      20880495
'#  File Load+Run: '    'test.tbl3'     '  Time: '      1.109813        '   Size:'      20824589
'#  File Load+Run: '    'test.tbl4'     '  Time: '      1.582231        '   Size:'      20827788

Bytecode files:

'#  File Load+Run: '    'test.out1'     '  Time: '      0.379144        '   Size:'      17521943
'#  File Load+Run: '    'test.out2'     '  Time: '      0.95122 '   Size:'      17519371
'#  File Load+Run: '    'test.out3'     '  Time: '      2.954594        '   Size:'      18333035
'#  File Load+Run: '    'test.out4'     '  Time: '      7.236041        '   Size:'      18327247

Oddly the first file is faster in bytecode (which is what I expected for all of them).

Can this be clearly explained? I have searched in many places (here specifically) and could not find any similar discussions (let me know if there are - please).