R 340 chars
cgc<-function(i="in.txt",x=100){
require(simecol)
z<-file("in.txt", "rb")
y<-matrix(data=NA,nrow=40,ncol=80)
for(i in seq(40)){
for(j in seq(80)){
y[i,j]<-ifelse(readChar(z,1) == "X",1,0)
}
readChar(z,3)
}
close(z)
init(conway) <- y
times(conway)<-1:x
o<-as.data.frame(out(sim(conway))[[100]])
write.table(o, "out.txt", sep="", row.names=FALSE, col.names=FALSE)
}
cgc()
I feel it's slightly cheating to have an add in package that does the actual automata for you, but I'm going with it cos I still had to thrash around with matricies and stuff to read in the file with 'X' instead of 1.
This is my first 'code golf', interesting....