Contigs can be short for one of the two reasons: 1) a contig connects to too few contigs; 2) a contig connects to too many contigs. The first thing is to check which is the case. For your data, it is more likely that 2) is happening when the assembler is picking up strain differences. Then you need an assembler that can aggressively prune error/variant-containing subgraphs. What assembler are you using? SPAdes is always a good start for small genomes. Velvet and my fermi-lite might be worth trying.