Better xz compression

Borrowed from downstream, we can be a little bit smarter
on how we compress modules.

parallel_xz.sh

@@ -0,0 +1,26 @@
+#!/bin/sh
+# Reads filenames on stdin, xz-compresses each in place.
+# Not optimal for "compress relatively few, large files" scenario!
+
+# How many xz's to run in parallel:
+procgroup=""
+while test "$#" != 0; do
+	# Get it from -jNUM
+	N="${1#-j}"
+	if test "$N" = "$1"; then
+		# Not -j<something> - warn and ignore
+		echo "parallel_xz: warning: unrecognized argument: '$1'"
+	else
+		procgroup="$N"
+	fi
+	shift
+done
+
+# If told to use only one cpu:
+test "$procgroup" || exec xargs -r xz
+test "$procgroup" = 1 && exec xargs -r xz
+
+# xz has some startup cost. If files are really small,
+# this cost might be significant. To combat this,
+# process several files (in sequence) by each xz process via -n 16:
+exec xargs -r -n 16 -P $procgroup xz