And because some of you may not have the determination to follow the link to the science20.com site, I post here some of the findings:

I decided to put together a simple program to extract winning probabilities with the above strategy. The program simulates the outcome of reds, blacks, and neutral numbers in 10,000,000 sessions of up to 200 wheel turns -the typical duration of a night-long game. The user may define the number of fiches after which he or she leaves the table (less than 200, of course, since we can win at most one fiche per turn of the wheel with our strategy), and the maximum bet allowed. Below I detail some results.

If the table has a maximum bet of 100 fiches, we get:

max 200 rolls, quit at +10: won 93.3435%, avg win= -9.2293

max 200 rolls, quit at +20: won 87.1386%, avg win=-17.4491

max 200 rolls, quit at +50: won 71.2252%, avg win=-36.6622

max 200 rolls, quit at +100: won 56.9104%, avg win=-51.7284

With a maximum bet of 1000 fiches, we get instead:

max 200 rolls, quit at +10: won 99.1284%, avg win= -9.2005

max 200 rolls, quit at +20: won 98.2628%, avg win=-18.2627

max 200 rolls, quit at +50: won 95.7134%, avg win=-44.6285

max 200 rolls, quit at +100: won 91.6532%, avg win=-78.0571

And with a maximum bet of 10000 fiches (if you can afford it!), we get:

max 200 rolls, quit at +10: won 99.876%, avg win=-11.8301

max 200 rolls, quit at +20: won 99.751%, avg win=-23.5798

max 200 rolls, quit at +50: won 99.379%, avg win=-59.0183

max 200 rolls, quit at +100: won 97.935%, avg win=-107.22

This graph shows the distribution of the longest streak of losses. This is what determines the long negative tail in the previous graph: