MongoDB disk read performance is very low - mongodb

Question

I have a MongoDB instance with a database and a collection with 80GB of data. The number of documents inside is about 4M with a comparatively large document size of about 20kB on average. Among other more elementary stuff, each documents contains one list of 1024 elements and also 3-4 lists of 200 numbers.
I perform a simple batch find query over a properly indexed string field ('isbn'), intending to get 5000 documents (projected on relevant part) in one batch. For this, I use the $in operator:
rows = COLLECTION.find({"isbn": {"$in": candidate_isbns}}, {"_id": 0,
"isbn": 1,
"other_stuff": 1})
The IXSCAN stage works properly as intended. Since the corresponding documents, however, are not within the WiredTiger cache yet (and probably never will be for my limited 32GB RAM), the data has to be read from disk during FETCH in most cases. (Unfortunately, "other_stuff" is too heavy to yield for an index that could cover this query.)
The SSD attached to my virtual cloud machine has a read performance of about 90MB/s, which is not great, but should be sufficient for now. However, when I monitor the disk read speed (via iostats, for example) the speed during the query goes down to roughly 3MB/s, which seems to be very poor. I can verify this poor behaviour by checking the profiler output (MongoDB seems to split the 5000 in further batches, so I show only the output for a sub-batch of 2094):
{
"op" : "getmore",
"ns" : "data.metadata",
"command" : {
"getMore" : NumberLong(7543502234201790529),
"collection" : "metadata",
"lsid" : {
"id" : UUID("2f410f2d-2f74-4d3a-9041-27c4ddc51bd2")
},
"$db" : "data"
},
"originatingCommand" : {
"$truncated" : "{ find: \"metadata\", filter: { isbn: { $in: [ \"9783927781313\", ..."
},
"cursorid" : NumberLong(7543502234201790529),
"keysExamined" : 4095,
"docsExamined" : 2095,
"numYield" : 803,
"nreturned" : 2094,
"locks" : {
"ReplicationStateTransition" : {
"acquireCount" : {
"w" : NumberLong(805)
}
},
"Global" : {
"acquireCount" : {
"r" : NumberLong(805)
}
},
"Database" : {
"acquireCount" : {
"r" : NumberLong(804)
}
},
"Collection" : {
"acquireCount" : {
"r" : NumberLong(804)
}
},
"Mutex" : {
"acquireCount" : {
"r" : NumberLong(1)
}
}
},
"flowControl" : {},
"storage" : {
"data" : {
"bytesRead" : NumberLong(65454770),
"timeReadingMicros" : NumberLong(21386543)
}
},
"responseLength" : 16769511,
"protocol" : "op_msg",
"millis" : 21745,
"planSummary" : "IXSCAN { isbn: 1 }",
"execStats" : {
"stage" : "PROJECTION_SIMPLE",
"nReturned" : 2196,
"executionTimeMillisEstimate" : 21126,
"works" : 4288,
"advanced" : 2196,
"needTime" : 2092,
"needYield" : 0,
"saveState" : 817,
"restoreState" : 817,
"isEOF" : 0,
"transformBy" : {},
"inputStage" : {
"stage" : "FETCH",
"nReturned" : 2196,
"executionTimeMillisEstimate" : 21116,
"works" : 4288,
"advanced" : 2196,
"needTime" : 2092,
"needYield" : 0,
"saveState" : 817,
"restoreState" : 817,
"isEOF" : 0,
"docsExamined" : 2196,
"alreadyHasObj" : 0,
"inputStage" : {
"stage" : "IXSCAN",
"nReturned" : 2196,
"executionTimeMillisEstimate" : 531,
"works" : 4288,
"advanced" : 2196,
"needTime" : 2092,
"needYield" : 0,
"saveState" : 817,
"restoreState" : 817,
"isEOF" : 0,
"keyPattern" : {
"isbn" : 1.0
},
"indexName" : "isbn_1",
"isMultiKey" : false,
"multiKeyPaths" : {
"isbn" : []
},
"isUnique" : true,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 2,
"direction" : "forward",
"indexBounds" : {
"isbn" : [
"[\"9780230391451\", \"9780230391451\"]",
"[\"9780230593206\", \"9780230593206\"]",
... ]
},
"keysExamined" : 4288,
"seeks" : 2093,
"dupsTested" : 0,
"dupsDropped" : 0
}
}
},
"ts" : ISODate("2022-01-24T07:57:12.132Z"),
"client" : "my_ip",
"allUsers" : [
{
"user" : "myUser",
"db" : "data"
}
],
"user" : "myUser#data"
}
By looking at the ratio of bytesRead vs. timeReadingMicros, this poor read speed of about 3MB/s can be confirmed, indeed.
My question: Why does this degradation of speed take place? Is it pathological, so that I need to do further investigations, or is it the expected behaviour, given the data setup above?
Any help is highly appreciated!