Those matrix sizes won't cause any problems. However, just so you know, if you were using double floating point elements (64 bits) the Mega only does 32-bit floating point. You will have to write whatever matrix operations are needed yourself.
That code is written for the Microsoft or Borland C/C++ compilers and uses a lot of other support libraries. I doubt that there's any way to get it to compile as-is on an Arduino. If you are only going to use a small number of matrix operations it would be easier to write them yourself - perhaps extracting some of the code from that file. Which Arduino do you have and what size of matrices are you going to use?
You don't have to convert them consecutively. There is a SKIP ROM command which you can issue on the bus which tells all of the DS18B20 to execute the next command without first checking for their own address. Then you just issue the temperature conversion command and wait for 750ms after which they will all have finished their conversion and you can read each one. Also, the 750ms delay is only required if you use the default 12 bit precision (to a sixteenth of a degree). If you set the devices to use 9 bit precision (to half a degree) the required delay is only about 95 milliseconds.