The Shortt-Synchronome clock [0] was the most accurate pendulum clock, known to have an error of less than 1 second per year. It was used in 1926 to detect tiny seasonal changes in the Earth's rotation rate. An evaluation of the clock in 1984 revealed it was even more accurate than thought --- it was in fact accurate to within 1 second per 12 years, the discrepancy being due to "the slight changes in gravity due to tidal distortions in the solid Earth caused by the gravity of the Sun and Moon."
Harrison's claim of 1 second error in 100 days is clearly in the same order of magnitude as the Shortt-Synchronome's known error rate of 1 second per year in 1926, and thus would have been also affected by the tiny seasonal changes in the Earth's rotation rate.
If the main purpose of the clock was to measure longitude by observing the position of the stars, "only as accurate as the Earth's rotation is consistent" is as good as you could possibly need.
However, the situations are actually pretty different. If you look into the study of the Shortt-Synchronome clock, you'll see it's actually talking about the gravitational effects of the sun and moon on pendulums. The Harrison clocks had no pendulums.
The Shortt-Synchronome is still a pretty amazing clock, though.
> The Harrison clocks had no pendulums.
The clock referred to in the article does have a pendulum.
For those who might be wondering why it's not every 24 hours, or 86400 seconds, look up "Sidereal Day"[0][1][2].
[0] https://www.google.co.uk/search?q=sidereal+day
[1] http://simple.wikipedia.org/wiki/Sidereal_day
[2] http://en.wikipedia.org/wiki/Sidereal_time