We recently performed a HiSeq 2000 PE-50 run for a customer. She had 15 samples of a species of invertebrate. These samples were each run on four lanes of a flowcell. Each sample had at least 80 million reads, with a mean quality score of no less than 36.7.
However, on cycle 14 of lane 1, for the top surface, there was an instrument malfunction where communication with the instrument camera was interrupted. As a result, all of the reads for that cycle of that lane were reported as "N." Therefore 1/8th of the reads for each sample have a value of "N" at cycle 14.
The customer believes that this missing cycle will cause problems with de novo transcriptome assembly. I've done plenty of genome assembly, but I don't have enough experience in transcriptome assembly to confidently disagree. Intuitively it seems that this missing base will be inconsequential to the final assembly.