It is not a power issue since I have looked at the signal over the oscilloscope. It is neither a problem of the pull up resistor. [...] I am pretty much sure is a software problem( the 1-wire library).
I'm not sure how you can conclude that it's not a hardware issue. You can't just continue to add devices to a 1-wire network and expect them all to perform as if they were the only devices on the bus. Maxim publishes an application note (AN148 - ""Guidelines for Reliable 1-Wire Networks") that discusses the issues people encounter when trying to build medium-sized and large 1-wire networks. It's readable and seems to contain a lot of good advice. Two things, in particular, that would probably be helpful would be to power the devices directly and to split your network up into two or more separate, smaller 1-Wire busses.
Personally, I don't think there's bug in the OneWire library that could be responsible for your problems. On the other hand, it may be possible to improve the reliability of your existing circuit by using the library differently. For example, you can either tell all the devices to start a temperature conversion at once, or you can address them individually and ask them to do their conversions one at a time. Having them do all their conversions at once is going to draw much more power from the bus, and so I'd avoid doing that. I guess a similar issue arises with the command to have devices report their addresses, but I'm not sure whether there's an alternative there. All that said, my feeling is that you will probably need to (at a minimum) split up your network into smaller, separately powered networks if you want to support that many devices.