Demonstable problem, Parallels Desktop (build 1848.2) and OS X 10.4.7 (build 8J2135a)

Here is a repeatable experiment that inidcates a problem between Parallels Desktop (build 1848.2) and OS X 10.4.7 (build 8J2135a) on my MacBook Pro (17"). The version of OS X is what gets downloaded via Software Update. The version of Parllels is from the CD I bought at a retail store.

The URLs used below have been discovered by repeated experiment to diagnose this problem and appear to quickly demonstrate the problem, but are by no means unique. At this point, Apple have replaced the logic card, so we can be reasonable sure that this is not a hardward problem.

Test 1:
With no Parallels installed (or with Parallels uninstalled followed by system restart);
Using the Ethernet port;
use Safari to hit, http://www.apple.com/imac
Observe no apparent distortion of the graphic.

Download a dmg file for Spinnn from http://www.apple.com/downloads/macosx/games/... (You may have to search for the file, but I chose this one because it is relatively small, and seems particularly problematic)
Observe that the dmg file mounts correctly.
Delete the dmg file.

Test 2:
Install Parallels. In my case:
kextstat | grep parallel
97 0 0x3cdd7000 0xe000 0xd000 com.parallels.kext.hypervisor (2.2) <11 6 5 4 3 2>
98 0 0x3cea1000 0x5000 0x4000 com.parallels.kext.vmmain (2.2) <11 6 5 4 3 2>
99 0 0x3cea6000 0x5000 0x4000 com.parallels.kext.Pvsnet (2.2) <5 4 3 2>
100 0 0x3cd19000 0x3000 0x2000 com.parallels.kext.Pvsvnic (2.2) <35 4 3>

use Safari, empty the cache and to hit, http://www.apple.com/imac. You may have to repeat this sequence a number of times, but in short order, observe distortion on the graphic in the page.

Download a dmg file for Spinnn from http://www.apple.com/downloads/macosx/games/...
Observe that attempts to mount the dmg result in the error message "codec overrun"
Delete the dmg file.

I can uninstall Parallels, restart, and repeat Test 1 and everything is fine. Repeat Test 2, and communication is not reliable.

The problem only seems to afflict the Ethernet port, since if I switch to an Airport connection, communications appears reliable during Test 2.