Let me get this straight...
privador,
Say you walk up just as the table opens for the day, so that the discard rack contains no cards: all six decks are being shuffled inside the machine.
You start playing. After two decks (call them decks A and B) have been played, let's say the HiLo runnning count is +16 (lots of small cards came out in decks A and B). Since the discard rack contains only two decks (A on the bottom and B on the top), the dealer leaves them in the discard rack.
The game continues. After another deck (deck C) goes by, the discard rack now contains three decks (A, B, and now C, from bottom to top). Thus, the dealer decides to return some of the cards to the machine. He takes the top deck from the discards (deck C) and returns them to the machine, so the discard rack once again contains just decks A and B.
The game continues again. After another deck (deck C') goes by, the discard rack now contains three decks (A, B, and now C', from bottom to top). Thus, the dealer decides to return some of the cards to the machine. He takes the top deck from the discards (deck C') and returns them to the machine, so the discard rack once again contains just decks A and B.
Now let's stop and analyze this situation. The discard rack always contains decks A and B: they are never returned to the shuffling machine.
Thus, if these two decks contain 16 more small cards than big cards (as in the example above), that means that the TC for the remaining four decks inside the machine is +16/4 = +4, which means that instead of a roughly 0.5% disadvantage, you have a roughly +1.5% advantage.
On the other hand, if these decks contain more big cards than small cards, then your disadvantage will be even worse than 0.5%.
The key, of course, is to know the count for the first two decks. If you're present while they're dealt, then you know their count immediately. If you approach a game in progress, where decks A and B are already fixed in the discard rack, then you have to estimate the count.
To estimate the TC, wait for the dealer to return some of the discards to the machine. Keep track of the RC for deck C (in other words, for the first deck out of the machine). Repeat for the next four or five deck C's, then average the results: this should be close to the actual TC for the four decks remaining in the machine.
If the approximate TC is +3 or higher, sit down and play! Your should have an advantage of at least 1%. While you're playing, keep checking the RC for each deck C, so ensure that your approximation is accurate.
Hope this helps!
Dog Hand