Are you using the sample endpoint, or the filter?
The basics here are that on average about 500 million Tweets are posted per day; so (roughly) up to 350,000 per minute; at 1% that’s around 3500 per minute, so more likely to be around 200,000 in an hour, but of course it will vary through a day, so I’d expect 100,000 to be in a reasonable range. “at any given time” would refer to the specific second(s) during which you’re connected to the stream.
If you’re using the filter / track, then you’re actually listening to all of the Tweets matching that filter which can peak at 1% of the firehose. So if you’re listening to #worldcup on a regular day, you’d “probably” get all the matching Tweets, since it’s unlikely that 1% of all traffic on Twitter is related to that event; but during that event happening (for example), the volume may be much higher, so you might miss some Tweets.