There may be other solutions, but it's fairly easy to accomplish using a PythonCaller, something like:

class FeatureCreator(object):
    def __init__(self):
        pass
    def input(self,feature):
        f = open("input_file.txt", "r")
        chunk = 100
        while True:
            data = f.read(chunk)
            if not data:
                break
            newFeature = fmeobjects.FMEFeature()
            newFeature.setAttribute('chunk' data)
            self.pyoutput(newFeature)
    def close(self):
        pass

Will return one feature with an attribute chunk for each chunk of 100 bytes in the file input_file.txt

i would attempt this with first getting the file into an attribute with AttributeFileReader then making a list from the record layout with AttributeSplitter.

class FeatureCreator(object):
    def __init__(self):
        pass
    def input(self,feature):
        f = open("input_file.txt", "r")
        chunk = 100
        while True:
            data = f.read(chunk)
            if not data:
                break
            newFeature = fmeobjects.FMEFeature()
            newFeature.setAttribute('chunk' data)
            self.pyoutput(newFeature)
    def close(self):
        pass

Here i was avoiding mentioning Python and David has no qualms!  The struct module is a candidate too ;-)

It's late Friday, my qualms have officially left the building ;-)

This is a good solution, given that the input file isn't very (very) large.

Thanks both for your suggestions. This was partly exploratory to see if FME could be used for this particular job. I'll try make sure I come back and post which solution was used and accept the relevant answer.