There are many tools to perform gap filling using Illumina short reads, for example "GapFiller: a de novo assembly approach to fill the gap within paired reads" or "Toward almost closed genomes with GapFiller". There are also some tools like GAPresolution that can help to perform local re-assemblies using 454 reads. We used GAPresolution but it is not a very good software, it is useful only in some specific situations.
Take a look at the PRICE software from the DeRisi lab. Its meant to do something very similar. http://derisilab.ucsf.edu/index.php?page=software
You could also look at SSPACE (http://www.baseclear.com/landingpages/basetools-a-wide-range-of-bioinformatics-solutions/sspacev12/, ATLAS tools (http://www.hgsc.bcm.tmc.edu/content/bcm-hgsc-software, and SCARPA (http://compbio.cs.toronto.edu/hapsembler/scarpa.html.
See the PAGIT protocol: http://www.sanger.ac.uk/resources/software/pagit/
In particular, take a look at the IMAGE tool: http://genomebiology.com/2010/11/4/R41
Also SOAPdenovo has ha function for scaffolding. Not sure about ABYSS
Here there is a useful explanation of several tools.
I could be wrong, but the above answers to your hypothetical scenario appear to miss the point that you aren't interested in assembling the full genome, just the 100 kb part you're interested in. I suggest the following algorithm:
1. Start with the initial assembly C0 of the contigs you have identified as overlapping your region of interest, and the set S of reads those contigs contain. Let C = C0.
2. Repeat:
a. Identify paired-end reads (not in C) for which one or both ends align within, or extending, contigs in C.
b. Identify unpaired reads that align extending these new paired-end reads.
c. Construct a new assembly C' from C and the new reads identified in (a) and (b).
d. Trim C' so it does not extend more than 100 kb to either end of C0. Set C = C'.
e. Let S' denote the reads that contribute to C'. If S' does not contain any reads not present in S, stop. Otherwise, Set S = S'.
3. If you don't have a complete assembly of the region of interest, generate an STS for each end of each contig, probe a library for clones including these STSes, subclone these clones into a paired-end sequencing vector, and generate paired-end reads for this library; then try steps (1) and (2) again, adding these new sequencing reads to what you had before.
4. If your average sequencing depth for the region of interest exceeds 25 or so without filling all gaps, it is likely that the remaining gaps represent sequences that are not getting cloned in your sequencing vectors. Try different sequencing vectors.
