For human and mouse ChIP-Seq experiments, we highly recommend at least 10 million analysis ready reads per sample (i.e. after low quality and duplicate removal). The read number may be less for smaller genomes.
Usually, NGS is done on pools of samples. Hence a lane can contain a mixture of libraries which need to be extracted into their corresponding samples. We use illumina's CASAVA software for most experiments, unless there are customized indexes and adapters present. The default output are gzipped files in the FASTQ format. These files are readily compatible with most NGS data analysis software currently out there.
At the McDermott Center, we are most careful about the quality of data generated by the sequencers. The first check is to make sure we have enough sequencing for the analysis. If samples do not pass the threshold required for the number of reads, they will be resequenced. Other parameters include % Passing Filter, Mean Quality Score and % of >=Q30 Bases to name a few. A detailed summary of the demultiplex stats used for initial Quality Accessing is available online.