This might explain many problems I have seen on a machine.

You need to tweak the /proc/sys/vm/bdflush parameters.

Currently on a heavy machine I use at work I have:

$ cat /proc/sys/vm/bdflush\n70      500    0       0       5000    5000    60      40      0

Your machine prolly looks like this:

$ cat /proc/sys/vm/bdflush\n50      500     0       0       500     3000    80      50      0

A good line for machine with HUGE amounts of Memory would be something

90      2000     0       0       5000    10000   100     50      0

or in /etc/sysctl.conf the line would be:

# Sets the File Systems to be more apropos for heavy IO and HUGE Memory\nvm.bdflush = 90 2000 0 0 5000 10000 100 50 0

There are other pieces that probably would help as well.

Edit:
Matter of fact I have found many pieces that are set too low for RedHat Enterprise Linux when you have 2+ Processors, gobs of memory and are IOWAIT bound. Here is a Good set of additions for you /etc/sysctl.conf:

# Sets the File Systems to be more\n# apropos for heavy IO and HUGE Memory\n# As RH set default are to low for Mega Systems\nvm.bdflush = 90 2000 0 0 5000 10000 100 50 0\nvm.kswapd = 1024 64 16\nvm.page-cluster = 5\nvm.inactive_clean_percent = 100\nvm.pagecache = 2 10 20\n\n# Controls low_latency for the Network driver\n# Being a GB Enet\nnet.ipv4.tcp_low_latency = 1

Oh, BTW, you probably won't REALLY have to do these Barry... Most of these setting are based on recs for Oracle and some for Appservers. Since this machine you have is a combo of everything... makes sense to do em.

Also, look at fs.lease-break-time the default == 45 seconds. (located /proc/sys/fs/lease-break-time. I don't know if you want to change this... but this maybe your wait period or maybe not.

What it comes down to, the machine sometimes has to grind away at the file because it is not cached. For the second wc -l it is in cache.

Since this machine is the one you described here, you are probably getting into the point were having 10GB of Memory is a bad thing, as it has to clear out cached items and put in the new stuff... of course you'd have to be caching at about 9GB, buffering at about 360MB and amoutn Free being in the sub 100MB free, like 40MB free.

Might also want to start thinking about changing the way the machine uses cacheing, and force it to use more memory for buffering as well.

$ cat /proc/sys/vm/bdflush
70      500    0       0       5000    5000    60      40      0

$ cat /proc/sys/vm/bdflush
50      500     0       0       500     3000    80      50      0

90      2000     0       0       5000    10000   100     50      0

# Sets the File Systems to be more apropos for heavy IO and HUGE Memory
vm.bdflush = 90 2000 0 0 5000 10000 100 50 0

$ cat /proc/sys/vm/bdflush
70      500    0       0       5000    5000    60      40      0

$ cat /proc/sys/vm/bdflush
50      500     0       0       500     3000    80      50      0

90      2000     0       0       5000    10000   100     50      0

# Sets the File Systems to be more apropos for heavy IO and HUGE Memory
vm.bdflush = 90 2000 0 0 5000 10000 100 50 0

$ cat /proc/sys/vm/bdflush
70      500    0       0       5000    5000    60      40      0

$ cat /proc/sys/vm/bdflush
50      500     0       0       500     3000    80      50      0

90      2000     0       0       5000    10000   100     50      0

# Sets the File Systems to be more apropos for heavy IO and HUGE Memory
vm.bdflush = 90 2000 0 0 5000 10000 100 50 0