Yes, you have to access each field by indexing each array, and so multiple iterations over each array-field.

for doc in timestampCursor:
    sysstat = doc['sysstat']
    for ts in sysstat['host']['statistics']['timestamp']:
        for cpu in ts['cpu-load-all']['cpu']:
            usr = cpu['usr']
            # Now, sum or average the 'usr' values, or whatever
            # you intend to do.

Alternatively, to aggregate the data server-side, you can use $unwind with $sum or $average or some other grouping operation with the MongoDB Aggregation Framework.

HOST_USAGE.aggregate([{
    '$match': {'sysstat.host.nodename': 1}
}, {
    # Rename the field for brevity.
    '$project': {'ts': '$sysstat.host.statistics.timestamp'}
}, {
    '$unwind': '$ts'
}, {
    '$unwind': '$ts.cpu-load-all.cpu'
}, {
    '$group': {
        '_id': 0,
        'all-usr': {'$sum': '$ts.cpu-load-all.cpu.usr'}
    }
}])))