Without testing further; it could be a problem that comes from the datatypes you use to store time.

Try using long for processing instead of int.

a__g, millis() is returning an int anyway, and it would take more than a month to overflow it...

outtaspace, I am suspecting that the Java HotSpot Just-in-Time compiler (or even the regular compiler!) optimizes your simple (if not simplistic) code, thus giving very good results, but not being representative of real speed of real code doing real, useful operations...
That's the inconvenience with micro-benchmarks, they often are faulty in their protocol of test.

So if I wanted to test it out what kinda of things would i have to do?

Btw the processing code was done in the processing editor and the C++ code was done in G++.

Both on my linux netbook. Would any of this change anything and if so what would i have to do to have a balanced test?

Also if the millis() returns an int it overflows at slightly less than a month 24.8551 days or double that if it returns an unsigned int.

After a bit of research I'm a bit curious about JIT. Does on wiki it says that it transfers code to machine code so it runs faster. Would this mean that it is faster?

Because for the work I'll be doing it'll mostly be small games and so forth. The best options I've found for solid coding and graphics are either XNA C# or processing. However I was trying to find out the actual power differences between C++, C# and processing to find out which was the most viable alternative. C# was my initial choice as it was the easiest, C++ was my next choice because it was multiplication and possibly more powerful and processing was my last choice as it was bar far the simplest had native svg support and so on. I wanted to do a quick speed test of all 3 languages to find out which was the best for my applications. I expected to see a small loss in the speed with processing and when I was getting up to 10X the speed I was pretty surprised.

A more realistic test, yet still simple to do, is to create an array of int to simulate a 800x800 (for example) picture: Processing allows to access such array (loadPixels(), pixels[]) to manipulate directly the pixels.
Create such array, fill it with random data, do a double loop (x, y, idx = x + y * width) to iterate on it, extract the R, G, B components with binary operators, change the values (reduce them by 10 % for example), merge back the channels and put back to the array.
Such test has the advantages to be realistic (lot of Processing sketches do that), and to exercise several parts of the language (array access, binary and arithmetic operations).

Another test would consist in creating a bunch of simple objects (say a ball, to remain classical) and put them in an array list. After filling it, the objects will be dequeued (thus, released), and other will be queued.
It will test object creation and destruction, and collection manipulation.

These tests must be run a number of times, to let the JIT to kick in.

The second test you proposed confused me slightly. I always thought there was no way to dequeue or destroy anything in java cause the garbage collector does what it wants. Also, kinda off topic, are there any pointers or references in processing like in C++ and C#? Have you ever tried any tests like this or felt that processing lacked speed in any cases? For example I found that running 100 squares and 100 text calls made my tablet's opengl crash.

Another thing, apart from PShapes is there any other things that should be done to optimize 2d graphics and text? Or should i start a new thread for new questions?

" 'are there any pointers or references in processing like in C++ and C#?"
Not really. In Java, you use references, implicitly.
For example, if I do:

1. Clss c = new Clss();
2. ArrayList<Clss> al = new ArrayList<Clss>();
3. al.add(c);

the array list contains not the object itself, but a reference to it.
If you do al.remove(0); it will remove the reference to the object and if it goes out of scope, effectively it will be garbage collected.