ParallelTable - performance problem

I noticed some issue while making parallel computations with Table loop.

I noticed that when one of kernels finishes its part of job, it stops working and only rest of kernels work. But when calculating huge tables it is potential loss of performance at the last rows of computation (I often end up having working only ONE kernel at last minutes of calculations). Is there any way to make all the parallel kernels work till the end? (So when one kernel finishes its job, it for example takes part of other rows to calculate on itself?) Thank you in advance for help!