Update:
Just doubled the software buffer values in the HardwareSerial.cpp (tuxduino was referring in his 1st post). Now it works.
As this can be intuitively ok for my main program, which e.g. has some if clauses during Serial.read(), I cannot understand why for the simple example I posted the buffer reaches values >68 (Especially because I also discovered, that changing the serial buffer iterativley to 9600 didn"t change anything). So am I already doing something "inefficient" in my example above (except from the print at the end, which doesn"t influence the buffer..).
What bad impact (RAM Size, speed in general) does the enlarging of the buffer has?
I see doubling the buffer size as a (temporary workaround) so still would be very happy to have some ideas regarding my original problem.