Just out of curiosity, I decided to see how much I had to worry about optimization in Lua. I downloaded a Lua binary from http://code.google.com/p/luaforwindows/ and ran a few tests.

Fresh boot, AMD dual core 2.6ghz. In case you're curious, that's a pretty old computer; it was at the price-performance sweet spot when I bought it, but parts are going to start failing soon.

local time = os.clock(); for i=1, 10^7 do if i==0 then break end end; print(os.clock()-time)
0.422

Note that this basic loop consists of 1 increment, 2 compares, and 1 jump. Every loop that follows is going to be the same, except sometimes they have 1 less compare (the if in the middle of the loop). I can run it more than 20 million times in a second. If I was doing this in ToME, I wouldn't notice any difference in framerate until I did it around 2*10^4 times every frame. That would drop my framerate from 30 to 29.

local time = os.clock(); local p; for i=1, 10^7 do p = math.sqrt(i) end; print(os.clock()-time)
1.766
Takes about four times longer to do a square root loop. I could only do 5000 square roots every frame before affecting Lua framerate.

local time = os.clock(); local p; for i=1, 10^7 do p = i^0.5 end; print(os.clock()-time)
1.11
Huh, that's funny. For the non math inclined, i^0.5 is the exact same thing as sqrt(i). I repeated this several times to be sure-- sure enough, ^0.5 is faster. And, I might add, easier to type, and easier on the eyes. So adjust that square estimate to 9000 a frame until we affect framerate.

local time = os.clock(); local p; for i=1, 10^7 do p = math.sin(i) end; print(os.clock()-time)
2.281
Slowest yet. Looks like I could get away with 4000 sines. I'm going to assume tan and cos work at similar speed (maybe not a safe assumption in light of the surprising square root discovery!)

local time = os.clock(); local p; for i=1, 10^7 do p = math.exp(i) end; print(os.clock()-time)
2.484
About the same as trig. Can't imagine wanting to do a lot of this.

local time = os.clock()local p = "test"; for i=1, 10^7 do if p~="test" then break end end; print(os.clock()-time)
0.422
Comparison of strings isn't any slower than comparison of numbers. That's because of how Lua handles strings. p is a pointer to a string; "test" is a pointer to the exact same string.

TL;DR: Unless you're writing display code or pathfinding code, there is almost nothing you can do that will noticeably impact the performance of ToME. Don't worry about speed-- worry about having your code do what it's supposed to do, and do it clearly.