I don't see why not, but you need to try. The starting point is make.input.format, which will be called like this:
make.input.format(format = function(con) {...} , mode="text", streaming.format="org.apache.hadoop.mapreduce.lib.input.NLineInputFormat")
The format argument is a function accepting an open connection as argument and returning a key-value pair, so it could be something like
function(con){line = readLines(con); keyval(NULL, strsplit(line, pattern = some.separator)[[1]])}
I don't know what the exact format is, so details may vary. The only problem is that there is no way to change the number of lines that this format will read in one step, it will be the default of one. Because of hadoop streaming limitations, the only way to pass arguments to the input format constructor is through the jobconf, and it doesn't look from the docs that this format reads anything from the jobconf, but I may be wrong. So it will work at the default of one line at a time. I hope that works for you
Antonio